2 * Copyright © 2011,2012 Google, Inc.
4 * This is part of HarfBuzz, a text shaping library.
6 * Permission is hereby granted, without written agreement and without
7 * license or royalty fees, to use, copy, modify, and distribute this
8 * software and its documentation for any purpose, provided that the
9 * above copyright notice and the following two paragraphs appear in
10 * all copies of this software.
12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
18 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
19 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
20 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
21 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
24 * Google Author(s): Behdad Esfahbod
27 #include "hb-ot-shape-complex-indic-private.hh"
28 #include "hb-ot-layout-private.hh"
30 /* buffer var allocations */
31 #define indic_category() complex_var_u8_0() /* indic_category_t */
32 #define indic_position() complex_var_u8_1() /* indic_position_t */
40 #define IN_HALF_BLOCK(u, Base) (((u) & ~0x7Fu) == (Base))
42 #define IS_DEVA(u) (IN_HALF_BLOCK (u, 0x0900u))
43 #define IS_BENG(u) (IN_HALF_BLOCK (u, 0x0980u))
44 #define IS_GURU(u) (IN_HALF_BLOCK (u, 0x0A00u))
45 #define IS_GUJR(u) (IN_HALF_BLOCK (u, 0x0A80u))
46 #define IS_ORYA(u) (IN_HALF_BLOCK (u, 0x0B00u))
47 #define IS_TAML(u) (IN_HALF_BLOCK (u, 0x0B80u))
48 #define IS_TELU(u) (IN_HALF_BLOCK (u, 0x0C00u))
49 #define IS_KNDA(u) (IN_HALF_BLOCK (u, 0x0C80u))
50 #define IS_MLYM(u) (IN_HALF_BLOCK (u, 0x0D00u))
51 #define IS_SINH(u) (IN_HALF_BLOCK (u, 0x0D80u))
52 #define IS_KHMR(u) (IN_HALF_BLOCK (u, 0x1780u))
55 #define MATRA_POS_LEFT(u) POS_PRE_M
56 #define MATRA_POS_RIGHT(u) ( \
57 IS_DEVA(u) ? POS_AFTER_SUB : \
58 IS_BENG(u) ? POS_AFTER_POST : \
59 IS_GURU(u) ? POS_AFTER_POST : \
60 IS_GUJR(u) ? POS_AFTER_POST : \
61 IS_ORYA(u) ? POS_AFTER_POST : \
62 IS_TAML(u) ? POS_AFTER_POST : \
63 IS_TELU(u) ? (u <= 0x0C42u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
64 IS_KNDA(u) ? (u < 0x0CC3u || u > 0xCD6u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
65 IS_MLYM(u) ? POS_AFTER_POST : \
66 IS_SINH(u) ? POS_AFTER_SUB : \
67 IS_KHMR(u) ? POS_AFTER_POST : \
68 /*default*/ POS_AFTER_SUB \
70 #define MATRA_POS_TOP(u) ( /* BENG and MLYM don't have top matras. */ \
71 IS_DEVA(u) ? POS_AFTER_SUB : \
72 IS_GURU(u) ? POS_AFTER_POST : /* Deviate from spec */ \
73 IS_GUJR(u) ? POS_AFTER_SUB : \
74 IS_ORYA(u) ? POS_AFTER_MAIN : \
75 IS_TAML(u) ? POS_AFTER_SUB : \
76 IS_TELU(u) ? POS_BEFORE_SUB : \
77 IS_KNDA(u) ? POS_BEFORE_SUB : \
78 IS_SINH(u) ? POS_AFTER_SUB : \
79 IS_KHMR(u) ? POS_AFTER_POST : \
80 /*default*/ POS_AFTER_SUB \
82 #define MATRA_POS_BOTTOM(u) ( \
83 IS_DEVA(u) ? POS_AFTER_SUB : \
84 IS_BENG(u) ? POS_AFTER_SUB : \
85 IS_GURU(u) ? POS_AFTER_POST : \
86 IS_GUJR(u) ? POS_AFTER_POST : \
87 IS_ORYA(u) ? POS_AFTER_SUB : \
88 IS_TAML(u) ? POS_AFTER_POST : \
89 IS_TELU(u) ? POS_BEFORE_SUB : \
90 IS_KNDA(u) ? POS_BEFORE_SUB : \
91 IS_MLYM(u) ? POS_AFTER_POST : \
92 IS_SINH(u) ? POS_AFTER_SUB : \
93 IS_KHMR(u) ? POS_AFTER_POST : \
94 /*default*/ POS_AFTER_SUB \
97 static inline indic_position_t
98 matra_position (hb_codepoint_t u, indic_position_t side)
102 case POS_PRE_C: return MATRA_POS_LEFT (u);
103 case POS_POST_C: return MATRA_POS_RIGHT (u);
104 case POS_ABOVE_C: return MATRA_POS_TOP (u);
105 case POS_BELOW_C: return MATRA_POS_BOTTOM (u);
111 * This is a hack for now. We should move this data into the main Indic table.
112 * Or completely remove it and just check in the tables.
114 static const hb_codepoint_t ra_chars[] = {
115 0x0930u, /* Devanagari */
116 0x09B0u, /* Bengali */
117 0x09F0u, /* Bengali */
118 0x0A30u, /* Gurmukhi */ /* No Reph */
119 0x0AB0u, /* Gujarati */
121 0x0BB0u, /* Tamil */ /* No Reph */
122 0x0C30u, /* Telugu */ /* Reph formed only with ZWJ */
123 0x0CB0u, /* Kannada */
124 0x0D30u, /* Malayalam */ /* No Reph, Logical Repha */
126 0x0DBBu, /* Sinhala */ /* Reph formed only with ZWJ */
128 0x179Au, /* Khmer */ /* No Reph, Visual Repha */
132 is_ra (hb_codepoint_t u)
134 for (unsigned int i = 0; i < ARRAY_LENGTH (ra_chars); i++)
135 if (u == ra_chars[i])
141 is_one_of (const hb_glyph_info_t &info, unsigned int flags)
143 /* If it ligated, all bets are off. */
144 if (_hb_glyph_info_ligated (&info)) return false;
145 return !!(FLAG_SAFE (info.indic_category()) & flags);
149 is_joiner (const hb_glyph_info_t &info)
151 return is_one_of (info, JOINER_FLAGS);
155 is_consonant (const hb_glyph_info_t &info)
157 return is_one_of (info, CONSONANT_FLAGS);
161 is_halant_or_coeng (const hb_glyph_info_t &info)
163 return is_one_of (info, HALANT_OR_COENG_FLAGS);
167 set_indic_properties (hb_glyph_info_t &info)
169 hb_codepoint_t u = info.codepoint;
170 unsigned int type = hb_indic_get_categories (u);
171 indic_category_t cat = (indic_category_t) (type & 0x7Fu);
172 indic_position_t pos = (indic_position_t) (type >> 8);
179 /* The following act more like the Bindus. */
180 if (unlikely (hb_in_range (u, 0x0953u, 0x0954u)))
182 /* The following act like consonants. */
183 else if (unlikely (hb_in_ranges (u, 0x0A72u, 0x0A73u,
186 /* TODO: The following should only be allowed after a Visarga.
187 * For now, just treat them like regular tone marks. */
188 else if (unlikely (hb_in_range (u, 0x1CE2u, 0x1CE8u)))
190 /* TODO: The following should only be allowed after some of
191 * the nasalization marks, maybe only for U+1CE9..U+1CF1.
192 * For now, just treat them like tone marks. */
193 else if (unlikely (u == 0x1CEDu))
195 /* The following take marks in standalone clusters, similar to Avagraha. */
196 else if (unlikely (hb_in_ranges (u, 0xA8F2u, 0xA8F7u,
201 ASSERT_STATIC ((int) INDIC_SYLLABIC_CATEGORY_AVAGRAHA == OT_Symbol);
203 else if (unlikely (u == 0x17C6u)) cat = OT_N; /* Khmer Bindu doesn't like to be repositioned. */
204 else if (unlikely (hb_in_range (u, 0x2010u, 0x2011u)))
205 cat = OT_PLACEHOLDER;
206 else if (unlikely (u == 0x25CCu)) cat = OT_DOTTEDCIRCLE;
210 * Re-assign position.
213 if ((FLAG_SAFE (cat) & CONSONANT_FLAGS))
219 else if (cat == OT_M)
221 pos = matra_position (u, pos);
223 else if ((FLAG_SAFE (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Symbol))))
228 if (unlikely (u == 0x0B01u)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */
232 info.indic_category() = cat;
233 info.indic_position() = pos;
237 * Things above this line should ideally be moved to the Indic table itself.
242 * Indic configurations. Note that we do not want to keep every single script-specific
243 * behavior in these tables necessarily. This should mainly be used for per-script
244 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and
245 * only one script has an exception, that one script can be if'ed directly in the code,
246 * instead of adding a new flag in these structs.
249 enum base_position_t {
251 BASE_POS_LAST_SINHALA,
254 enum reph_position_t {
255 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN,
256 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB,
257 REPH_POS_AFTER_SUB = POS_AFTER_SUB,
258 REPH_POS_BEFORE_POST = POS_BEFORE_POST,
259 REPH_POS_AFTER_POST = POS_AFTER_POST,
260 REPH_POS_DONT_CARE = POS_RA_TO_BECOME_REPH
263 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */
264 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */
265 REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */
266 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */
269 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */
270 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */
272 struct indic_config_t
276 hb_codepoint_t virama;
277 base_position_t base_pos;
278 reph_position_t reph_pos;
279 reph_mode_t reph_mode;
280 blwf_mode_t blwf_mode;
283 static const indic_config_t indic_configs[] =
285 /* Default. Should be first. */
286 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
287 {HB_SCRIPT_DEVANAGARI,true, 0x094Du,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
288 {HB_SCRIPT_BENGALI, true, 0x09CDu,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
289 {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
290 {HB_SCRIPT_GUJARATI, true, 0x0ACDu,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
291 {HB_SCRIPT_ORIYA, true, 0x0B4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
292 {HB_SCRIPT_TAMIL, true, 0x0BCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
293 {HB_SCRIPT_TELUGU, true, 0x0C4Du,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY},
294 {HB_SCRIPT_KANNADA, true, 0x0CCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY},
295 {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST},
296 {HB_SCRIPT_SINHALA, false,0x0DCAu,BASE_POS_LAST_SINHALA,
297 REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST},
298 {HB_SCRIPT_KHMER, false,0x17D2u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST},
307 struct feature_list_t {
309 hb_ot_map_feature_flags_t flags;
312 static const feature_list_t
317 * These features are applied in order, one at a time, after initial_reordering.
319 {HB_TAG('n','u','k','t'), F_GLOBAL},
320 {HB_TAG('a','k','h','n'), F_GLOBAL},
321 {HB_TAG('r','p','h','f'), F_NONE},
322 {HB_TAG('r','k','r','f'), F_GLOBAL},
323 {HB_TAG('p','r','e','f'), F_NONE},
324 {HB_TAG('b','l','w','f'), F_NONE},
325 {HB_TAG('a','b','v','f'), F_NONE},
326 {HB_TAG('h','a','l','f'), F_NONE},
327 {HB_TAG('p','s','t','f'), F_NONE},
328 {HB_TAG('v','a','t','u'), F_GLOBAL},
329 {HB_TAG('c','j','c','t'), F_GLOBAL},
330 {HB_TAG('c','f','a','r'), F_NONE},
333 * These features are applied all at once, after final_reordering.
334 * Default Bengali font in Windows for example has intermixed
335 * lookups for init,pres,abvs,blws features.
337 {HB_TAG('i','n','i','t'), F_NONE},
338 {HB_TAG('p','r','e','s'), F_GLOBAL},
339 {HB_TAG('a','b','v','s'), F_GLOBAL},
340 {HB_TAG('b','l','w','s'), F_GLOBAL},
341 {HB_TAG('p','s','t','s'), F_GLOBAL},
342 {HB_TAG('h','a','l','n'), F_GLOBAL},
343 /* Positioning features, though we don't care about the types. */
344 {HB_TAG('d','i','s','t'), F_GLOBAL},
345 {HB_TAG('a','b','v','m'), F_GLOBAL},
346 {HB_TAG('b','l','w','m'), F_GLOBAL},
350 * Must be in the same order as the indic_features array.
377 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */
381 setup_syllables (const hb_ot_shape_plan_t *plan,
383 hb_buffer_t *buffer);
385 initial_reordering (const hb_ot_shape_plan_t *plan,
387 hb_buffer_t *buffer);
389 final_reordering (const hb_ot_shape_plan_t *plan,
391 hb_buffer_t *buffer);
393 clear_syllables (const hb_ot_shape_plan_t *plan,
395 hb_buffer_t *buffer);
398 collect_features_indic (hb_ot_shape_planner_t *plan)
400 hb_ot_map_builder_t *map = &plan->map;
402 /* Do this before any lookups have been applied. */
403 map->add_gsub_pause (setup_syllables);
405 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
406 /* The Indic specs do not require ccmp, but we apply it here since if
407 * there is a use of it, it's typically at the beginning. */
408 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
412 map->add_gsub_pause (initial_reordering);
413 for (; i < INDIC_BASIC_FEATURES; i++) {
414 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
415 map->add_gsub_pause (NULL);
417 map->add_gsub_pause (final_reordering);
418 for (; i < INDIC_NUM_FEATURES; i++) {
419 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
422 map->add_global_bool_feature (HB_TAG('c','a','l','t'));
423 map->add_global_bool_feature (HB_TAG('c','l','i','g'));
425 map->add_gsub_pause (clear_syllables);
429 override_features_indic (hb_ot_shape_planner_t *plan)
431 /* Uniscribe does not apply 'kern' in Khmer. */
432 if (hb_options ().uniscribe_bug_compatible)
434 switch ((hb_tag_t) plan->props.script)
436 case HB_SCRIPT_KHMER:
437 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL);
442 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
446 struct would_substitute_feature_t
448 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
450 zero_context = zero_context_;
451 map->get_stage_lookups (0/*GSUB*/,
452 map->get_feature_stage (0/*GSUB*/, feature_tag),
456 inline bool would_substitute (const hb_codepoint_t *glyphs,
457 unsigned int glyphs_count,
458 hb_face_t *face) const
460 for (unsigned int i = 0; i < count; i++)
461 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context))
467 const hb_ot_map_t::lookup_map_t *lookups;
472 struct indic_shape_plan_t
476 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
478 hb_codepoint_t glyph = virama_glyph;
479 if (unlikely (virama_glyph == (hb_codepoint_t) -1))
481 if (!config->virama || !font->get_nominal_glyph (config->virama, &glyph))
483 /* Technically speaking, the spec says we should apply 'locl' to virama too.
484 * Maybe one day... */
486 /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph
487 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */
488 (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph;
495 const indic_config_t *config;
498 hb_codepoint_t virama_glyph;
500 would_substitute_feature_t rphf;
501 would_substitute_feature_t pref;
502 would_substitute_feature_t blwf;
503 would_substitute_feature_t pstf;
505 hb_mask_t mask_array[INDIC_NUM_FEATURES];
509 data_create_indic (const hb_ot_shape_plan_t *plan)
511 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t));
512 if (unlikely (!indic_plan))
515 indic_plan->config = &indic_configs[0];
516 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++)
517 if (plan->props.script == indic_configs[i].script) {
518 indic_plan->config = &indic_configs[i];
522 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2');
523 indic_plan->virama_glyph = (hb_codepoint_t) -1;
525 /* Use zero-context would_substitute() matching for new-spec of the main
526 * Indic scripts, and scripts with one spec only, but not for old-specs.
527 * The new-spec for all dual-spec scripts says zero-context matching happens.
529 * However, testing with Malayalam shows that old and new spec both allow
530 * context. Testing with Bengali new-spec however shows that it doesn't.
531 * So, the heuristic here is the way it is. It should *only* be changed,
532 * as we discover more cases of what Windows does. DON'T TOUCH OTHERWISE.
534 bool zero_context = !indic_plan->is_old_spec && plan->props.script != HB_SCRIPT_MALAYALAM;
535 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
536 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
537 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);
538 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context);
540 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++)
541 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ?
542 0 : plan->map.get_1_mask (indic_features[i].tag);
548 data_destroy_indic (void *data)
553 static indic_position_t
554 consonant_position_from_face (const indic_shape_plan_t *indic_plan,
555 const hb_codepoint_t consonant,
556 const hb_codepoint_t virama,
559 /* For old-spec, the order of glyphs is Consonant,Virama,
560 * whereas for new-spec, it's Virama,Consonant. However,
561 * some broken fonts (like Free Sans) simply copied lookups
562 * from old-spec to new-spec without modification.
563 * And oddly enough, Uniscribe seems to respect those lookups.
564 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds
565 * base at 0. The font however, only has lookups matching
566 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec
567 * table). As such, we simply match both sequences. Seems
569 hb_codepoint_t glyphs[3] = {virama, consonant, virama};
570 if (indic_plan->blwf.would_substitute (glyphs , 2, face) ||
571 indic_plan->blwf.would_substitute (glyphs+1, 2, face))
573 if (indic_plan->pstf.would_substitute (glyphs , 2, face) ||
574 indic_plan->pstf.would_substitute (glyphs+1, 2, face))
576 if (indic_plan->pref.would_substitute (glyphs , 2, face) ||
577 indic_plan->pref.would_substitute (glyphs+1, 2, face))
583 enum syllable_type_t {
592 #include "hb-ot-shape-complex-indic-machine.hh"
596 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED,
598 hb_font_t *font HB_UNUSED)
600 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
601 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
603 /* We cannot setup masks here. We save information about characters
604 * and setup masks later on in a pause-callback. */
606 unsigned int count = buffer->len;
607 hb_glyph_info_t *info = buffer->info;
608 for (unsigned int i = 0; i < count; i++)
609 set_indic_properties (info[i]);
613 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
614 hb_font_t *font HB_UNUSED,
617 find_syllables (buffer);
621 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
623 int a = pa->indic_position();
624 int b = pb->indic_position();
626 return a < b ? -1 : a == b ? 0 : +1;
632 update_consonant_positions (const hb_ot_shape_plan_t *plan,
636 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
638 if (indic_plan->config->base_pos != BASE_POS_LAST)
641 hb_codepoint_t virama;
642 if (indic_plan->get_virama_glyph (font, &virama))
644 hb_face_t *face = font->face;
645 unsigned int count = buffer->len;
646 hb_glyph_info_t *info = buffer->info;
647 for (unsigned int i = 0; i < count; i++)
648 if (info[i].indic_position() == POS_BASE_C)
650 hb_codepoint_t consonant = info[i].codepoint;
651 info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face);
658 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */
661 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
664 unsigned int start, unsigned int end)
666 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
667 hb_glyph_info_t *info = buffer->info;
670 /* 1. Find base consonant:
672 * The shaping engine finds the base consonant of the syllable, using the
673 * following algorithm: starting from the end of the syllable, move backwards
674 * until a consonant is found that does not have a below-base or post-base
675 * form (post-base forms have to follow below-base forms), or that is not a
676 * pre-base reordering Ra, or arrive at the first consonant. The consonant
677 * stopped at will be the base.
679 * o If the syllable starts with Ra + Halant (in a script that has Reph)
680 * and has more than one consonant, Ra is excluded from candidates for
684 unsigned int base = end;
685 bool has_reph = false;
688 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
689 * and has more than one consonant, Ra is excluded from candidates for
690 * base consonants. */
691 unsigned int limit = start;
692 if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE &&
693 indic_plan->mask_array[RPHF] &&
696 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) ||
697 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ)
700 /* See if it matches the 'rphf' feature. */
701 hb_codepoint_t glyphs[3] = {info[start].codepoint,
702 info[start + 1].codepoint,
703 indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ?
704 info[start + 2].codepoint : 0};
705 if (indic_plan->rphf.would_substitute (glyphs, 2, face) ||
706 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT &&
707 indic_plan->rphf.would_substitute (glyphs, 3, face)))
710 while (limit < end && is_joiner (info[limit]))
715 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha)
718 while (limit < end && is_joiner (info[limit]))
724 switch (indic_plan->config->base_pos)
728 /* -> starting from the end of the syllable, move backwards */
729 unsigned int i = end;
730 bool seen_below = false;
733 /* -> until a consonant is found */
734 if (is_consonant (info[i]))
736 /* -> that does not have a below-base or post-base form
737 * (post-base forms have to follow below-base forms), */
738 if (info[i].indic_position() != POS_BELOW_C &&
739 (info[i].indic_position() != POS_POST_C || seen_below))
744 if (info[i].indic_position() == POS_BELOW_C)
747 /* -> or that is not a pre-base reordering Ra,
749 * IMPLEMENTATION NOTES:
751 * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped
752 * by the logic above already.
755 /* -> or arrive at the first consonant. The consonant stopped at will
761 /* A ZWJ after a Halant stops the base search, and requests an explicit
763 * A ZWJ before a Halant, requests a subjoined form instead, and hence
764 * search continues. This is particularly important for Bengali
765 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */
767 info[i].indic_category() == OT_ZWJ &&
768 info[i - 1].indic_category() == OT_H)
775 case BASE_POS_LAST_SINHALA:
777 /* Sinhala base positioning is slightly different from main Indic, in that:
778 * 1. Its ZWJ behavior is different,
779 * 2. We don't need to look into the font for consonant positions.
785 /* Find the last base consonant that is not blocked by ZWJ. If there is
786 * a ZWJ right before a base consonant, that would request a subjoined form. */
787 for (unsigned int i = limit; i < end; i++)
788 if (is_consonant (info[i]))
790 if (limit < i && info[i - 1].indic_category() == OT_ZWJ)
796 /* Mark all subsequent consonants as below. */
797 for (unsigned int i = base + 1; i < end; i++)
798 if (is_consonant (info[i]))
799 info[i].indic_position() = POS_BELOW_C;
805 /* The first consonant is always the base. */
807 assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA);
812 /* Mark all subsequent consonants as below. */
813 for (unsigned int i = base + 1; i < end; i++)
814 if (is_consonant (info[i]))
815 info[i].indic_position() = POS_BELOW_C;
820 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
821 * and has more than one consonant, Ra is excluded from candidates for
824 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */
825 if (has_reph && base == start && limit - base <= 2) {
826 /* Have no other consonant, so Reph is not formed and Ra becomes base. */
832 /* 2. Decompose and reorder Matras:
834 * Each matra and any syllable modifier sign in the cluster are moved to the
835 * appropriate position relative to the consonant(s) in the cluster. The
836 * shaping engine decomposes two- or three-part matras into their constituent
837 * parts before any repositioning. Matra characters are classified by which
838 * consonant in a conjunct they have affinity for and are reordered to the
839 * following positions:
841 * o Before first half form in the syllable
842 * o After subjoined consonants
843 * o After post-form consonant
844 * o After main consonant (for above marks)
846 * IMPLEMENTATION NOTES:
848 * The normalize() routine has already decomposed matras for us, so we don't
849 * need to worry about that.
853 /* 3. Reorder marks to canonical order:
855 * Adjacent nukta and halant or nukta and vedic sign are always repositioned
856 * if necessary, so that the nukta is first.
858 * IMPLEMENTATION NOTES:
860 * We don't need to do this: the normalize() routine already did this for us.
864 /* Reorder characters */
866 for (unsigned int i = start; i < base; i++)
867 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position());
870 info[base].indic_position() = POS_BASE_C;
872 /* Mark final consonants. A final consonant is one appearing after a matra,
874 for (unsigned int i = base + 1; i < end; i++)
875 if (info[i].indic_category() == OT_M) {
876 for (unsigned int j = i + 1; j < end; j++)
877 if (is_consonant (info[j])) {
878 info[j].indic_position() = POS_FINAL_C;
884 /* Handle beginning Ra */
886 info[start].indic_position() = POS_RA_TO_BECOME_REPH;
888 /* For old-style Indic script tags, move the first post-base Halant after
891 * Reports suggest that in some scripts Uniscribe does this only if there
892 * is *not* a Halant after last consonant already (eg. Kannada), while it
893 * does it unconditionally in other scripts (eg. Malayalam). We don't
894 * currently know about other scripts, so we single out Malayalam for now.
897 * U+0C9A,U+0CCD,U+0C9A,U+0CCD
898 * With some versions of Lohit Kannada.
899 * https://bugs.freedesktop.org/show_bug.cgi?id=59118
901 * Malayalam test case:
902 * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D
903 * With lohit-ttf-20121122/Lohit-Malayalam.ttf
905 if (indic_plan->is_old_spec)
907 bool disallow_double_halants = buffer->props.script != HB_SCRIPT_MALAYALAM;
908 for (unsigned int i = base + 1; i < end; i++)
909 if (info[i].indic_category() == OT_H)
912 for (j = end - 1; j > i; j--)
913 if (is_consonant (info[j]) ||
914 (disallow_double_halants && info[j].indic_category() == OT_H))
916 if (info[j].indic_category() != OT_H && j > i) {
917 /* Move Halant to after last consonant. */
918 hb_glyph_info_t t = info[i];
919 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0]));
926 /* Attach misc marks to previous char to move with them. */
928 indic_position_t last_pos = POS_START;
929 for (unsigned int i = start; i < end; i++)
931 if ((FLAG_SAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS)))
933 info[i].indic_position() = last_pos;
934 if (unlikely (info[i].indic_category() == OT_H &&
935 info[i].indic_position() == POS_PRE_M))
938 * Uniscribe doesn't move the Halant with Left Matra.
939 * TEST: U+092B,U+093F,U+094DE
940 * We follow. This is important for the Sinhala
941 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
942 * where U+0DD9 is a left matra and U+0DCA is the virama.
943 * We don't want to move the virama with the left matra.
944 * TEST: U+0D9A,U+0DDA
946 for (unsigned int j = i; j > start; j--)
947 if (info[j - 1].indic_position() != POS_PRE_M) {
948 info[i].indic_position() = info[j - 1].indic_position();
952 } else if (info[i].indic_position() != POS_SMVD) {
953 last_pos = (indic_position_t) info[i].indic_position();
957 /* For post-base consonants let them own anything before them
958 * since the last consonant or matra. */
960 unsigned int last = base;
961 for (unsigned int i = base + 1; i < end; i++)
962 if (is_consonant (info[i]))
964 for (unsigned int j = last + 1; j < i; j++)
965 if (info[j].indic_position() < POS_SMVD)
966 info[j].indic_position() = info[i].indic_position();
968 } else if (info[i].indic_category() == OT_M)
974 /* Use syllable() for sort accounting temporarily. */
975 unsigned int syllable = info[start].syllable();
976 for (unsigned int i = start; i < end; i++)
977 info[i].syllable() = i - start;
979 /* Sit tight, rock 'n roll! */
980 hb_stable_sort (info + start, end - start, compare_indic_order);
981 /* Find base again */
983 for (unsigned int i = start; i < end; i++)
984 if (info[i].indic_position() == POS_BASE_C)
989 /* Things are out-of-control for post base positions, they may shuffle
990 * around like crazy. In old-spec mode, we move halants around, so in
991 * that case merge all clusters after base. Otherwise, check the sort
992 * order and merge as needed.
993 * For pre-base stuff, we handle cluster issues in final reordering.
995 * We could use buffer->sort() for this, if there was no special
996 * reordering of pre-base stuff happening later...
998 if (indic_plan->is_old_spec || end - base > 127)
999 buffer->merge_clusters (base, end);
1002 /* Note! syllable() is a one-byte field. */
1003 for (unsigned int i = base; i < end; i++)
1004 if (info[i].syllable() != 255)
1006 unsigned int max = i;
1007 unsigned int j = start + info[i].syllable();
1011 unsigned int next = start + info[j].syllable();
1012 info[j].syllable() = 255; /* So we don't process j later again. */
1016 buffer->merge_clusters (i, max + 1);
1020 /* Put syllable back in. */
1021 for (unsigned int i = start; i < end; i++)
1022 info[i].syllable() = syllable;
1025 /* Setup masks now */
1031 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++)
1032 info[i].mask |= indic_plan->mask_array[RPHF];
1035 mask = indic_plan->mask_array[HALF];
1036 if (!indic_plan->is_old_spec &&
1037 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST)
1038 mask |= indic_plan->mask_array[BLWF];
1039 for (unsigned int i = start; i < base; i++)
1040 info[i].mask |= mask;
1044 info[base].mask |= mask;
1046 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF];
1047 for (unsigned int i = base + 1; i < end; i++)
1048 info[i].mask |= mask;
1051 if (indic_plan->is_old_spec &&
1052 buffer->props.script == HB_SCRIPT_DEVANAGARI)
1054 /* Old-spec eye-lash Ra needs special handling. From the
1057 * "The feature 'below-base form' is applied to consonants
1058 * having below-base forms and following the base consonant.
1059 * The exception is vattu, which may appear below half forms
1060 * as well as below the base glyph. The feature 'below-base
1061 * form' will be applied to all such occurrences of Ra as well."
1063 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915
1064 * with Sanskrit 2003 font.
1066 * However, note that Ra,Halant,ZWJ is the correct way to
1067 * request eyelash form of Ra, so we wouldbn't inhibit it
1070 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915
1072 for (unsigned int i = start; i + 1 < base; i++)
1073 if (info[i ].indic_category() == OT_Ra &&
1074 info[i+1].indic_category() == OT_H &&
1076 info[i+2].indic_category() != OT_ZWJ))
1078 info[i ].mask |= indic_plan->mask_array[BLWF];
1079 info[i+1].mask |= indic_plan->mask_array[BLWF];
1083 unsigned int pref_len = 2;
1084 if (indic_plan->mask_array[PREF] && base + pref_len < end)
1086 /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */
1087 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) {
1088 hb_codepoint_t glyphs[2];
1089 for (unsigned int j = 0; j < pref_len; j++)
1090 glyphs[j] = info[i + j].codepoint;
1091 if (indic_plan->pref.would_substitute (glyphs, pref_len, face))
1093 for (unsigned int j = 0; j < pref_len; j++)
1094 info[i++].mask |= indic_plan->mask_array[PREF];
1096 /* Mark the subsequent stuff with 'cfar'. Used in Khmer.
1097 * Read the feature spec.
1098 * This allows distinguishing the following cases with MS Khmer fonts:
1099 * U+1784,U+17D2,U+179A,U+17D2,U+1782
1100 * U+1784,U+17D2,U+1782,U+17D2,U+179A
1102 if (indic_plan->mask_array[CFAR])
1103 for (; i < end; i++)
1104 info[i].mask |= indic_plan->mask_array[CFAR];
1111 /* Apply ZWJ/ZWNJ effects */
1112 for (unsigned int i = start + 1; i < end; i++)
1113 if (is_joiner (info[i])) {
1114 bool non_joiner = info[i].indic_category() == OT_ZWNJ;
1120 /* ZWJ/ZWNJ should disable CJCT. They do that by simply
1121 * being there, since we don't skip them for the CJCT
1122 * feature (ie. F_MANUAL_ZWJ) */
1124 /* A ZWNJ disables HALF. */
1126 info[j].mask &= ~indic_plan->mask_array[HALF];
1128 } while (j > start && !is_consonant (info[j]));
1133 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan,
1135 hb_buffer_t *buffer,
1136 unsigned int start, unsigned int end)
1138 /* We treat placeholder/dotted-circle as if they are consonants, so we
1139 * should just chain. Only if not in compatibility mode that is... */
1141 if (hb_options ().uniscribe_bug_compatible)
1143 /* For dotted-circle, this is what Uniscribe does:
1144 * If dotted-circle is the last glyph, it just does nothing.
1145 * Ie. It doesn't form Reph. */
1146 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE)
1150 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1154 initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
1156 hb_buffer_t *buffer,
1157 unsigned int start, unsigned int end)
1159 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
1160 switch (syllable_type)
1162 case vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */
1163 case consonant_syllable:
1164 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1167 case broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */
1168 case standalone_cluster:
1169 initial_reordering_standalone_cluster (plan, face, buffer, start, end);
1172 case symbol_cluster:
1173 case non_indic_cluster:
1179 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
1181 hb_buffer_t *buffer)
1183 /* Note: This loop is extra overhead, but should not be measurable. */
1184 bool has_broken_syllables = false;
1185 unsigned int count = buffer->len;
1186 hb_glyph_info_t *info = buffer->info;
1187 for (unsigned int i = 0; i < count; i++)
1188 if ((info[i].syllable() & 0x0F) == broken_cluster)
1190 has_broken_syllables = true;
1193 if (likely (!has_broken_syllables))
1197 hb_codepoint_t dottedcircle_glyph;
1198 if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph))
1201 hb_glyph_info_t dottedcircle = {0};
1202 dottedcircle.codepoint = 0x25CCu;
1203 set_indic_properties (dottedcircle);
1204 dottedcircle.codepoint = dottedcircle_glyph;
1206 buffer->clear_output ();
1209 unsigned int last_syllable = 0;
1210 while (buffer->idx < buffer->len && !buffer->in_error)
1212 unsigned int syllable = buffer->cur().syllable();
1213 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
1214 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
1216 last_syllable = syllable;
1218 hb_glyph_info_t ginfo = dottedcircle;
1219 ginfo.cluster = buffer->cur().cluster;
1220 ginfo.mask = buffer->cur().mask;
1221 ginfo.syllable() = buffer->cur().syllable();
1222 /* TODO Set glyph_props? */
1224 /* Insert dottedcircle after possible Repha. */
1225 while (buffer->idx < buffer->len && !buffer->in_error &&
1226 last_syllable == buffer->cur().syllable() &&
1227 buffer->cur().indic_category() == OT_Repha)
1228 buffer->next_glyph ();
1230 buffer->output_info (ginfo);
1233 buffer->next_glyph ();
1236 buffer->swap_buffers ();
1240 initial_reordering (const hb_ot_shape_plan_t *plan,
1242 hb_buffer_t *buffer)
1244 update_consonant_positions (plan, font, buffer);
1245 insert_dotted_circles (plan, font, buffer);
1247 foreach_syllable (buffer, start, end)
1248 initial_reordering_syllable (plan, font->face, buffer, start, end);
1252 final_reordering_syllable (const hb_ot_shape_plan_t *plan,
1253 hb_buffer_t *buffer,
1254 unsigned int start, unsigned int end)
1256 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
1257 hb_glyph_info_t *info = buffer->info;
1260 /* This function relies heavily on halant glyphs. Lots of ligation
1261 * and possibly multiplication substitutions happened prior to this
1262 * phase, and that might have messed up our properties. Recover
1263 * from a particular case of that where we're fairly sure that a
1264 * class of OT_H is desired but has been lost. */
1265 if (indic_plan->virama_glyph)
1267 unsigned int virama_glyph = indic_plan->virama_glyph;
1268 for (unsigned int i = start; i < end; i++)
1269 if (info[i].codepoint == virama_glyph &&
1270 _hb_glyph_info_ligated (&info[i]) &&
1271 _hb_glyph_info_multiplied (&info[i]))
1273 /* This will make sure that this glyph passes is_halant_or_coeng() test. */
1274 info[i].indic_category() = OT_H;
1275 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]);
1280 /* 4. Final reordering:
1282 * After the localized forms and basic shaping forms GSUB features have been
1283 * applied (see below), the shaping engine performs some final glyph
1284 * reordering before applying all the remaining font features to the entire
1288 bool try_pref = !!indic_plan->mask_array[PREF];
1290 /* Find base again */
1292 for (base = start; base < end; base++)
1293 if (info[base].indic_position() >= POS_BASE_C)
1295 if (try_pref && base + 1 < end)
1297 for (unsigned int i = base + 1; i < end; i++)
1298 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1300 if (!(_hb_glyph_info_substituted (&info[i]) &&
1301 _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1303 /* Ok, this was a 'pref' candidate but didn't form any.
1304 * Base is around here... */
1306 while (base < end && is_halant_or_coeng (info[base]))
1308 info[base].indic_position() = POS_BASE_C;
1315 /* For Malayalam, skip over unformed below- (but NOT post-) forms. */
1316 if (buffer->props.script == HB_SCRIPT_MALAYALAM)
1318 for (unsigned int i = base + 1; i < end; i++)
1320 while (i < end && is_joiner (info[i]))
1322 if (i == end || !is_halant_or_coeng (info[i]))
1324 i++; /* Skip halant. */
1325 while (i < end && is_joiner (info[i]))
1327 if (i < end && is_consonant (info[i]) && info[i].indic_position() == POS_BELOW_C)
1330 info[base].indic_position() = POS_BASE_C;
1335 if (start < base && info[base].indic_position() > POS_BASE_C)
1339 if (base == end && start < base &&
1340 is_one_of (info[base - 1], FLAG (OT_ZWJ)))
1343 while (start < base &&
1344 is_one_of (info[base], (FLAG (OT_N) | HALANT_OR_COENG_FLAGS)))
1348 /* o Reorder matras:
1350 * If a pre-base matra character had been reordered before applying basic
1351 * features, the glyph can be moved closer to the main consonant based on
1352 * whether half-forms had been formed. Actual position for the matra is
1353 * defined as “after last standalone halant glyph, after initial matra
1354 * position and before the main consonant”. If ZWJ or ZWNJ follow this
1355 * halant, position is moved after it.
1358 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */
1360 /* If we lost track of base, alas, position before last thingy. */
1361 unsigned int new_pos = base == end ? base - 2 : base - 1;
1363 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1364 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1365 * We want to position matra after them.
1367 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1369 while (new_pos > start &&
1370 !(is_one_of (info[new_pos], (FLAG (OT_M) | HALANT_OR_COENG_FLAGS))))
1373 /* If we found no Halant we are done.
1374 * Otherwise only proceed if the Halant does
1375 * not belong to the Matra itself! */
1376 if (is_halant_or_coeng (info[new_pos]) &&
1377 info[new_pos].indic_position() != POS_PRE_M)
1379 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1380 if (new_pos + 1 < end && is_joiner (info[new_pos + 1]))
1384 new_pos = start; /* No move. */
1387 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M)
1389 /* Now go see if there's actually any matras... */
1390 for (unsigned int i = new_pos; i > start; i--)
1391 if (info[i - 1].indic_position () == POS_PRE_M)
1393 unsigned int old_pos = i - 1;
1394 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */
1397 hb_glyph_info_t tmp = info[old_pos];
1398 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0]));
1399 info[new_pos] = tmp;
1401 /* Note: this merge_clusters() is intentionally *after* the reordering.
1402 * Indic matra reordering is special and tricky... */
1403 buffer->merge_clusters (new_pos, MIN (end, base + 1));
1408 for (unsigned int i = start; i < base; i++)
1409 if (info[i].indic_position () == POS_PRE_M) {
1410 buffer->merge_clusters (i, MIN (end, base + 1));
1419 * Reph’s original position is always at the beginning of the syllable,
1420 * (i.e. it is not reordered at the character reordering stage). However,
1421 * it will be reordered according to the basic-forms shaping results.
1422 * Possible positions for reph, depending on the script, are; after main,
1423 * before post-base consonant forms, and after post-base consonant forms.
1428 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then
1429 * we should only move it if the sequence ligated to the repha form.
1431 * - If repha is encoded separately and in the logical position, we should only
1432 * move it if it did NOT ligate. If it ligated, it's probably the font trying
1433 * to make it work without the reordering.
1435 if (start + 1 < end &&
1436 info[start].indic_position() == POS_RA_TO_BECOME_REPH &&
1437 ((info[start].indic_category() == OT_Repha) ^
1438 _hb_glyph_info_ligated_and_didnt_multiply (&info[start])))
1440 unsigned int new_reph_pos;
1441 reph_position_t reph_pos = indic_plan->config->reph_pos;
1443 assert (reph_pos != REPH_POS_DONT_CARE);
1445 /* 1. If reph should be positioned after post-base consonant forms,
1446 * proceed to step 5.
1448 if (reph_pos == REPH_POS_AFTER_POST)
1453 /* 2. If the reph repositioning class is not after post-base: target
1454 * position is after the first explicit halant glyph between the
1455 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ
1456 * are following this halant, position is moved after it. If such
1457 * position is found, this is the target position. Otherwise,
1458 * proceed to the next step.
1460 * Note: in old-implementation fonts, where classifications were
1461 * fixed in shaping engine, there was no case where reph position
1462 * will be found on this step.
1465 new_reph_pos = start + 1;
1466 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1469 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1471 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1472 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1478 /* 3. If reph should be repositioned after the main consonant: find the
1479 * first consonant not ligated with main, or find the first
1480 * consonant that is not a potential pre-base reordering Ra.
1482 if (reph_pos == REPH_POS_AFTER_MAIN)
1484 new_reph_pos = base;
1485 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN)
1487 if (new_reph_pos < end)
1491 /* 4. If reph should be positioned before post-base consonant, find
1492 * first post-base classified consonant not ligated with main. If no
1493 * consonant is found, the target position should be before the
1494 * first matra, syllable modifier sign or vedic sign.
1496 /* This is our take on what step 4 is trying to say (and failing, BADLY). */
1497 if (reph_pos == REPH_POS_AFTER_SUB)
1499 new_reph_pos = base;
1500 while (new_reph_pos < end &&
1501 !( FLAG_SAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD))))
1503 if (new_reph_pos < end)
1507 /* 5. If no consonant is found in steps 3 or 4, move reph to a position
1508 * immediately before the first post-base matra, syllable modifier
1509 * sign or vedic sign that has a reordering class after the intended
1510 * reph position. For example, if the reordering position for reph
1511 * is post-main, it will skip above-base matras that also have a
1512 * post-main position.
1516 /* Copied from step 2. */
1517 new_reph_pos = start + 1;
1518 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1521 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1523 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1524 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1530 /* 6. Otherwise, reorder reph to the end of the syllable.
1533 new_reph_pos = end - 1;
1534 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD)
1538 * If the Reph is to be ending up after a Matra,Halant sequence,
1539 * position it before that Halant so it can interact with the Matra.
1540 * However, if it's a plain Consonant,Halant we shouldn't do that.
1541 * Uniscribe doesn't do this.
1542 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D
1544 if (!hb_options ().uniscribe_bug_compatible &&
1545 unlikely (is_halant_or_coeng (info[new_reph_pos]))) {
1546 for (unsigned int i = base + 1; i < new_reph_pos; i++)
1547 if (info[i].indic_category() == OT_M) {
1558 buffer->merge_clusters (start, new_reph_pos + 1);
1559 hb_glyph_info_t reph = info[start];
1560 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0]));
1561 info[new_reph_pos] = reph;
1563 if (start < base && base <= new_reph_pos)
1569 /* o Reorder pre-base reordering consonants:
1571 * If a pre-base reordering consonant is found, reorder it according to
1572 * the following rules:
1575 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */
1577 for (unsigned int i = base + 1; i < end; i++)
1578 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1580 /* 1. Only reorder a glyph produced by substitution during application
1581 * of the <pref> feature. (Note that a font may shape a Ra consonant with
1582 * the feature generally but block it in certain contexts.)
1584 /* Note: We just check that something got substituted. We don't check that
1585 * the <pref> feature actually did it...
1587 * Reorder pref only if it ligated. */
1588 if (_hb_glyph_info_ligated_and_didnt_multiply (&info[i]))
1591 * 2. Try to find a target position the same way as for pre-base matra.
1592 * If it is found, reorder pre-base consonant glyph.
1594 * 3. If position is not found, reorder immediately before main
1598 unsigned int new_pos = base;
1599 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1600 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1601 * We want to position matra after them.
1603 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1605 while (new_pos > start &&
1606 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS)))
1609 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a
1610 * split matra, it should be reordered to *before* the left part of such matra. */
1611 if (new_pos > start && info[new_pos - 1].indic_category() == OT_M)
1613 unsigned int old_pos = i;
1614 for (unsigned int j = base + 1; j < old_pos; j++)
1615 if (info[j].indic_category() == OT_M)
1623 if (new_pos > start && is_halant_or_coeng (info[new_pos - 1]))
1625 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1626 if (new_pos < end && is_joiner (info[new_pos]))
1631 unsigned int old_pos = i;
1633 buffer->merge_clusters (new_pos, old_pos + 1);
1634 hb_glyph_info_t tmp = info[old_pos];
1635 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0]));
1636 info[new_pos] = tmp;
1638 if (new_pos <= base && base < old_pos)
1648 /* Apply 'init' to the Left Matra if it's a word start. */
1649 if (info[start].indic_position () == POS_PRE_M &&
1651 !(FLAG_SAFE (_hb_glyph_info_get_general_category (&info[start - 1])) &
1652 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))))
1653 info[start].mask |= indic_plan->mask_array[INIT];
1657 * Finish off the clusters and go home!
1659 if (hb_options ().uniscribe_bug_compatible)
1661 switch ((hb_tag_t) plan->props.script)
1663 case HB_SCRIPT_TAMIL:
1664 case HB_SCRIPT_SINHALA:
1668 /* Uniscribe merges the entire cluster... Except for Tamil & Sinhala.
1669 * This means, half forms are submerged into the main consonants cluster.
1670 * This is unnecessary, and makes cursor positioning harder, but that's what
1671 * Uniscribe does. */
1672 buffer->merge_clusters (start, end);
1680 final_reordering (const hb_ot_shape_plan_t *plan,
1681 hb_font_t *font HB_UNUSED,
1682 hb_buffer_t *buffer)
1684 unsigned int count = buffer->len;
1685 if (unlikely (!count)) return;
1687 foreach_syllable (buffer, start, end)
1688 final_reordering_syllable (plan, buffer, start, end);
1690 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
1691 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);
1696 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
1697 hb_font_t *font HB_UNUSED,
1698 hb_buffer_t *buffer)
1700 hb_glyph_info_t *info = buffer->info;
1701 unsigned int count = buffer->len;
1702 for (unsigned int i = 0; i < count; i++)
1703 info[i].syllable() = 0;
1708 decompose_indic (const hb_ot_shape_normalize_context_t *c,
1715 /* Don't decompose these. */
1716 case 0x0931u : return false;
1717 case 0x0B94u : return false;
1721 * Decompose split matras that don't have Unicode decompositions.
1724 case 0x0F77u : *a = 0x0FB2u; *b= 0x0F81u; return true;
1725 case 0x0F79u : *a = 0x0FB3u; *b= 0x0F81u; return true;
1726 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true;
1727 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true;
1728 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true;
1729 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true;
1730 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true;
1731 case 0x1925u : *a = 0x1920u; *b= 0x1923u; return true;
1732 case 0x1926u : *a = 0x1920u; *b= 0x1924u; return true;
1733 case 0x1B3Cu : *a = 0x1B42u; *b= 0x1B3Cu; return true;
1734 case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true;
1735 case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true;
1737 /* This one has no decomposition in Unicode, but needs no decomposition either. */
1738 /* case 0x0AC9u : return false; */
1739 case 0x0B57u : *a = no decomp, -> RIGHT; return true;
1740 case 0x1C29u : *a = no decomp, -> LEFT; return true;
1741 case 0xA9C0u : *a = no decomp, -> RIGHT; return true;
1742 case 0x111BuF : *a = no decomp, -> ABOVE; return true;
1746 if ((ab == 0x0DDAu || hb_in_range (ab, 0x0DDCu, 0x0DDEu)))
1749 * Sinhala split matras... Let the fun begin.
1751 * These four characters have Unicode decompositions. However, Uniscribe
1752 * decomposes them "Khmer-style", that is, it uses the character itself to
1753 * get the second half. The first half of all four decompositions is always
1756 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are
1757 * broken with Uniscribe. But we need to support them. As such, we only
1758 * do the Uniscribe-style decomposition if the character is transformed into
1759 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to
1760 * Unicode decomposition.
1762 * Note that we can't unconditionally use Unicode decomposition. That would
1763 * break some other fonts, that are designed to work with Uniscribe, and
1764 * don't have positioning features for the Unicode-style decomposition.
1768 * The Uniscribe behavior is now documented in the newly published Sinhala
1771 * http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping
1774 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data;
1776 hb_codepoint_t glyph;
1778 if (hb_options ().uniscribe_bug_compatible ||
1779 (c->font->get_nominal_glyph (ab, &glyph) &&
1780 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face)))
1782 /* Ok, safe to use Uniscribe-style decomposition. */
1789 return (bool) c->unicode->decompose (ab, a, b);
1793 compose_indic (const hb_ot_shape_normalize_context_t *c,
1798 /* Avoid recomposing split matras. */
1799 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
1802 /* Composition-exclusion exceptions that we want to recompose. */
1803 if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; }
1805 return (bool) c->unicode->compose (a, b, ab);
1809 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
1812 collect_features_indic,
1813 override_features_indic,
1816 NULL, /* preprocess_text */
1817 NULL, /* postprocess_glyphs */
1818 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
1822 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
1823 false, /* fallback_position */